Forecasting of the Urban Area State Using Convolutional Neural Networks

Abstract

Active development of modern cities requires not only efficient monitoring systems but furthermore forecasting systems that can predict future state of the urban area with high accuracy. In this work we present a method for urban area prediction based on geospatial activity of users in social network. One of the most popular social networks, Instagram, was taken as a source for spatial data and two large cities with different peculiarities of online activity-New York City, USA, and Saint Petersburg, Russia - were taken as target cities. We propose three different deep learning architectures that are able to solve a target problem and show that convolutional neural network based on three-dimensional convolution layers provides the best results with accuracy of 99%.